Research map / Probabilistic and scientific modeling / Variational and generative sampling
Bi-Level Doubly Variational Learning for Energy-Based Latent Variable Models
Ge Kan, Jinhu Lü, Tian Wang, Baochang Zhang, Aichun Zhu, Lei Huang, Guodong Guo, Hichem Snoussi
Check your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use the complete decoupled EBLVM with symmetric KL, rather than symmetric KL alone in a coupled EBLVM (ablation) | BiDVL · CIFAR-10 | FID: 32.10 -> 20.75 | Improved |
| set the basic importance ratio to 0.05 rather than 0.01 (ablation) | BiDVL · CIFAR-10 | FID: 22.62 -> 20.75 | Improved |
| increase the basic importance ratio from 0.05 to 0.5 or 1.0 (ablation) | BiDVL · CIFAR-10 | FID: 20.75 -> 26.82; 29.72 | Hurt |
| offset the opposite gradient terms (ablation) | BiDVL · CIFAR-10 | FID: 21.54 -> 20.75 | Improved |
| offset the opposite gradient terms (ablation) | BiDVL · CIFAR-10 | RMSE: 0.170 -> 0.168 | Improved |
| remove the opposite terms from the algorithm (ablation) | BiDVL · CIFAR-10 | FID: 20.75 -> 20.24 | Improved |
| remove the opposite terms from the algorithm (ablation) | BiDVL · CIFAR-10 | RMSE: 0.168 -> 0.175 | Hurt |
| remove the opposite terms from the algorithm (ablation) | BiDVL · CIFAR-10 training; SVHN and CelebA OOD | AUROC: 0.81 AUROC on SVHN; 0.75 AUROC on CelebA | Mixed |
| use the sigmoid estimate for the importance-ratio bias term instead of ignoring the bias term (ablation) | BiDVL · image generation | FID | Hurt |
| use a ResNet-based architecture (ablation) | BiDVL · CIFAR-10 | FID: 20.75 -> 16.37 | Improved |
Closest papers by meaning
- Bi-level Score Matching for Learning Energy-based Latent Variable Models (NeurIPS 2020 · similarity 0.90)
- Variational (Gradient) Estimate of the Score Function in Energy-based Latent Variable Models (ICML 2021 · similarity 0.88)
- Conjugate Energy-Based Models (ICML 2021 · similarity 0.86)
- Learning Equivariant Energy Based Models with Equivariant Stein Variational Gradient Descent (NeurIPS 2021 · similarity 0.85)
- Joint Training of Variational Auto-Encoder and Latent Energy-Based Model (CVPR 2020 · similarity 0.85)
- VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models (ICLR 2021 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
